OpenAI Preps 100+ Solutions to Open Math Problems, and Mathematicians Aren't Happy

OpenAI is preparing to release more than 100 new solutions to previously unsolved mathematics problems, a move that has drawn sharp criticism from mathematicians who describe the company's conduct as heavy-handed. The results, first reported by Wired, are expected to cover a range of open problems, though OpenAI has not said exactly which ones or how the work was produced. The dispute is less about whether the answers are correct than about how a commercial lab is inserting itself into a field with its own slow, peer-reviewed norms.

One mathematician told Wired there is "a perception of mobster behavior" from leading AI companies, a phrase that captures how some researchers view the way these firms announce results. The complaint is not that AI cannot contribute to mathematics — several systems have produced useful conjectures and proofs in recent years — but that splashy releases can bypass the usual channels of verification. Mathematicians typically publish through journals and arXiv preprints, where claims are checked, challenged and sometimes overturned over months or years.

OpenAI has pushed into scientific domains repeatedly as it looks to demonstrate the reasoning power of its models. Math is an attractive target because answers are objectively checkable and because strong performance signals capabilities that transfer to coding, engineering and other technical work. The company has not disclosed whether the 100-plus solutions were found by a model working alone, by a model paired with human researchers, or by some other method, and that ambiguity is part of what has irritated outside mathematicians.

The friction also reflects a structural imbalance. AI labs hold compute, funding and publicity reach that individual mathematicians and university departments cannot match, so when a lab announces a batch of results, the field has little leverage to slow it down or demand evidence first. If the solutions hold up, the researchers who posed those problems may see years of work resolved by a system they had no role in building. If some turn out to be wrong or trivial restatements, the cost falls on the mathematicians who have to sort through them.

For now, the key question is verification. OpenAI has not published a timeline for releasing the solutions, named the mathematicians involved, or explained how the results will be checked. Until independent researchers can examine the claims, the argument is mostly about process and posture rather than mathematical content. What happens next — whether OpenAI submits the work for ordinary peer review or simply posts it — will shape how much of the field treats the results as a contribution rather than a provocation.